Nonpher

Nonpher generates virtual libraries of hard-to-synthesize chemical structures to enable exploration of chemical space with a focus on synthetic feasibility for cheminformatics, drug discovery, and materials science.

Key Features:

• Molecular Morphing Algorithm: Iteratively generates new chemical structures via simple modifications such as the addition or removal of atoms or bonds.
• Machine Learning–based Library Construction: Leverages machine learning techniques to construct virtual libraries of hard-to-synthesize compounds.
• Optimization for Synthetic Feasibility: Produces structures balanced for complexity and synthetic challenge, explicitly optimizing for synthetic feasibility.
• Comparative Evaluation with SAscore and dense region (DR): Evaluated against SAscore and dense region (DR) datasets, with a random forest classifier trained on Nonpher-generated data outperforming models trained on SAscore and DR data.

Scientific Applications:

• Drug Discovery: Generates hard-to-synthesize compounds to expand exploration of novel chemical spaces for potential drug candidates.
• Cheminformatics Research: Provides datasets and insights into synthetic feasibility for development and benchmarking of predictive models.
• Materials Science: Enables exploration of structurally challenging compounds relevant to materials discovery and design.

Methodology:

Constructs virtual libraries by iteratively applying structural changes using a molecular morphing algorithm and machine learning, and trains/evaluates a random forest classifier on Nonpher-generated data versus SAscore and dense region (DR) datasets.